also experiment a bit with status register in log

* sht3x: add crc and status debug

* reduce delays, add sensor debug

* add full status output

fixing reset always keeping the last value, not the intended size 3... and above.
revert to the previous behaviour where the last element is preserved and
real `capacity` becomes `capacity + 1`. adjust methods accordingly.
and a slightly more verbose median funciton

amends 5fcac5d27f

generalize topic and value representation
breaking change for rpn, since the magnitude *may not* be indexed

also fixes debug logs, where we would always see units of the 1st
magnitude instead of the one being processed

keeping the last value will prevent us from updating
anything due to capacity check

also see #2519

see #2519

third time's a charm... reorganize external funcs to reference internal
ones, and there's no need for an extra check of address

flags as code blocks, typos

since we don't have multiple instances yet, move everything into the
existing sensor instance class. for the multi-device migration, we will
only need port and it's read timeout setting. address will be stored per
device instance, where the main one will work with the port

make sure we use clamp instead of constrain

resolves #2501

Co-Authored-by: Maxim Prokhorov <prokhorov.max@outlook.com>

and make sure we use it when trying to read/write something

make sure we don't just hope that pointer logic works

some swserial migrations
also make sure we use pstrs

simplify the internals and allow to use multiple instances

possibly fix comparison issue with raw int vs. float, when checking for
errors. division may not actually succeed

migrate to the new structure

stronger time types, overrides

init underlying lib object as a member
this might be changed to unique_ptr, in case we really want begin() to
be idempotent and being able to re-init things
(which might not be actually possible?)

structure, overrides, strong time types

actually use the analog sensor base to read
(and fixing yet another pinmode accident)

fixed structure, overrides. strong time types

updated structure, overrides. strong time type
testing timesource changes. this should work, but might need to tweak
things if it starts to fail b/c of flash code loads

conversion is not working for signed ints, only unsigned

do not force to implement constructor, just update vtable
continue testing with bme280 class, update structure

make sure there's an option for direct address setup
sensor updates are wip, since base sensor code is also changing

use bitset when locking, avoid manual index & mask calculation
update check() to use boolean logic instead of returning int status
remove wip address-range, just use the {begin, end} pair
uint8_t instead of unsigned char when dealing with data

Avoid confusion between ref and pointer when looping over sensors
Re-use helper class in the magnitude object as well

Remove custom code related to filter movement, just use unique_ptr
Plus, updates filter code to use stl vector where possible

Experimening with dynamic cls id. Perhaps this should be something
else... but, at least there are no longer any header-defined values.

preemptive fix for the issue resolved in the 3.1.0